Page MenuHomePhabricator
Feed Advanced Search

Fri, May 24

rampitec accepted D62401: [AMDGPU] Speed up live-in virtual register set computaion in GCNScheduleDAGMILive.

LGTM

Fri, May 24, 10:50 AM · Restricted Project
rampitec accepted D62355: AMDGPU: Activate all lanes when spilling CSR VGPR for SGPR spills.

LGTM

Fri, May 24, 10:49 AM
rampitec added a comment to D62401: [AMDGPU] Speed up live-in virtual register set computaion in GCNScheduleDAGMILive.

LGTM, but you need to split LiveInterval.h part in a parent review.

Fri, May 24, 10:38 AM · Restricted Project
rampitec accepted D62396: AMDGPU: Boost inline threshold with addrspacecasted alloca arguments.

LGTM

Fri, May 24, 8:50 AM

Thu, May 23

rampitec added a comment to D62355: AMDGPU: Activate all lanes when spilling CSR VGPR for SGPR spills.

Where do you get SGPR pair for saveexec if you need to spill an SGPR?

Thu, May 23, 5:05 PM
rampitec accepted D59990: AMDGPU. Divergence driven ISel. Assign register class for cross block values according to the divergence..

LGTM

Thu, May 23, 11:39 AM · Restricted Project
rampitec accepted D62310: AMDGPU: Correct maximum possible private allocation size.

LGTM

Thu, May 23, 11:12 AM

Wed, May 22

rampitec added a comment to D62134: MC: Allow getMaxInstLength to depend on the subtarget.

JFYI, you need to mention nsa in the commit description.

Wed, May 22, 8:34 AM
rampitec accepted D62134: MC: Allow getMaxInstLength to depend on the subtarget.

LGTM

Wed, May 22, 8:32 AM

Tue, May 21

rampitec accepted D62222: AMDGPU: Assume calls read exec.

LGTM

Tue, May 21, 3:21 PM
rampitec accepted D62217: AMDGPU: Assume call pseudos are convergent.

LGTM

Tue, May 21, 2:55 PM
rampitec committed rG44d17ca02e0b: Fix register coalescer failure to prune value (authored by rampitec).
Fix register coalescer failure to prune value
Tue, May 21, 12:32 PM
rampitec committed rL361293: Fix register coalescer failure to prune value.
Fix register coalescer failure to prune value
Tue, May 21, 12:31 PM
rampitec closed D62162: Fix register coalescer failure to prune value.
Tue, May 21, 12:31 PM · Restricted Project
rampitec accepted D62194: AMDGPU: Fix not marking new gfx10 SGPRs as CSRs.

LGTM

Tue, May 21, 10:46 AM
rampitec updated the diff for D62162: Fix register coalescer failure to prune value.

Use test further reduced by Matt.

Tue, May 21, 10:45 AM · Restricted Project

Mon, May 20

rampitec updated the diff for D62162: Fix register coalescer failure to prune value.

Reduced test.

Mon, May 20, 5:23 PM · Restricted Project
rampitec accepted D62134: MC: Allow getMaxInstLength to depend on the subtarget.

LGTM

Mon, May 20, 5:23 PM
rampitec added inline comments to D62162: Fix register coalescer failure to prune value.
Mon, May 20, 3:10 PM · Restricted Project
rampitec created D62162: Fix register coalescer failure to prune value.
Mon, May 20, 2:41 PM · Restricted Project
rampitec accepted D62161: AMDGPU: Force skip branches over calls.

LGTM

Mon, May 20, 2:38 PM
rampitec added inline comments to D62134: MC: Allow getMaxInstLength to depend on the subtarget.
Mon, May 20, 11:05 AM
rampitec added a comment to D59990: AMDGPU. Divergence driven ISel. Assign register class for cross block values according to the divergence..

LGTM. Let's finish with internal integration and testing before proceeding.

Mon, May 20, 10:51 AM · Restricted Project

Sat, May 18

rampitec accepted D62058: [AMDGPU] gfx1010 Avoid SMEM WAR hazard for some s_waitcnt values.

LGTM

Sat, May 18, 8:56 AM · Restricted Project

Fri, May 17

rampitec added inline comments to D62058: [AMDGPU] gfx1010 Avoid SMEM WAR hazard for some s_waitcnt values.
Fri, May 17, 9:46 AM · Restricted Project

Wed, May 15

rampitec accepted D61988: AMDGPU: Introduce TokenFactor for ABI register copies in call sequence.

LGTM

Wed, May 15, 8:57 PM
rampitec added inline comments to D59990: AMDGPU. Divergence driven ISel. Assign register class for cross block values according to the divergence..
Wed, May 15, 8:15 AM · Restricted Project

Tue, May 14

rampitec accepted D61901: AMDGPU: Assume xnack is enabled by default.

We need to emphasize this change also enables xnack by default on all other targets and probably need to be complimented with FE/RT changes.

Tue, May 14, 10:47 AM
rampitec accepted D61905: [AMDGPU][MC][GFX8][GFX9] Corrected predicate of v_*_co_u32 aliases.

LGTM

Tue, May 14, 10:43 AM · Restricted Project
rampitec added a comment to D61901: AMDGPU: Assume xnack is enabled by default.

Why do we have both xnack and no-xnack? I thought we are going to drop xnack feature in favor of no-xnack.

Tue, May 14, 10:33 AM
rampitec added a comment to D61905: [AMDGPU][MC][GFX8][GFX9] Corrected predicate of v_*_co_u32 aliases.

The comment does not belong here. Also please retitle.

Tue, May 14, 10:30 AM · Restricted Project
rampitec committed rG05791d90c916: [AMDGPU] Fixed handling of imemdiate i1 literals (authored by rampitec).
[AMDGPU] Fixed handling of imemdiate i1 literals
Tue, May 14, 9:17 AM
rampitec committed rL360689: [AMDGPU] Fixed handling of imemdiate i1 literals.
[AMDGPU] Fixed handling of imemdiate i1 literals
Tue, May 14, 9:17 AM
rampitec closed D61812: [AMDGPU] Fixed handling of imemdiate i1 literals.
Tue, May 14, 9:17 AM · Restricted Project
rampitec added inline comments to D61812: [AMDGPU] Fixed handling of imemdiate i1 literals.
Tue, May 14, 9:14 AM · Restricted Project
rampitec committed rG7b20032628c0: [AMDGPU] gfx1010 Strengthen some SMEM WAR hazard unit tests. NFC. (authored by rampitec).
[AMDGPU] gfx1010 Strengthen some SMEM WAR hazard unit tests. NFC.
Tue, May 14, 9:04 AM
rampitec committed rL360686: [AMDGPU] gfx1010 Strengthen some SMEM WAR hazard unit tests. NFC..
[AMDGPU] gfx1010 Strengthen some SMEM WAR hazard unit tests. NFC.
Tue, May 14, 9:01 AM
rampitec added a comment to D61905: [AMDGPU][MC][GFX8][GFX9] Corrected predicate of v_*_co_u32 aliases.
In D61905#1501536, @dp wrote:

SWDEV-188513 is P2. Could you correct the offending predicate by your next commit?
Or I can remove all changes except the predicate correction - that will not affect GFX10.

Tue, May 14, 8:57 AM · Restricted Project
rampitec updated the diff for D61812: [AMDGPU] Fixed handling of imemdiate i1 literals.

Changed return to true.

Tue, May 14, 8:48 AM · Restricted Project
rampitec requested changes to D61905: [AMDGPU][MC][GFX8][GFX9] Corrected predicate of v_*_co_u32 aliases.

GFX10 changes are not yet complete.

Tue, May 14, 8:45 AM · Restricted Project

Mon, May 13

rampitec committed rG91792f1b93e9: [AMDGPU] gfx1010 clang target (authored by rampitec).
[AMDGPU] gfx1010 clang target
Mon, May 13, 4:17 PM
rampitec committed rL360634: [AMDGPU] gfx1010 clang target.
[AMDGPU] gfx1010 clang target
Mon, May 13, 4:17 PM
rampitec committed rC360634: [AMDGPU] gfx1010 clang target.
[AMDGPU] gfx1010 clang target
Mon, May 13, 4:17 PM
rampitec closed D61875: [AMDGPU] gfx1010 clang target.
Mon, May 13, 4:17 PM · Restricted Project
rampitec added inline comments to D61528: Revert "[AMDGPU] Enable code selection using `s_mul_hi_u32`/`s_mul_hi_i32`.".
Mon, May 13, 3:13 PM
rampitec added inline comments to D61528: Revert "[AMDGPU] Enable code selection using `s_mul_hi_u32`/`s_mul_hi_i32`.".
Mon, May 13, 3:02 PM
rampitec committed rGec42fc177c1d: [AMDGPU] gfx1010 SearchableTableEmitter patch for NSA (authored by rampitec).
[AMDGPU] gfx1010 SearchableTableEmitter patch for NSA
Mon, May 13, 3:00 PM
rampitec committed rL360623: [AMDGPU] gfx1010 SearchableTableEmitter patch for NSA.
[AMDGPU] gfx1010 SearchableTableEmitter patch for NSA
Mon, May 13, 2:57 PM
rampitec closed D61868: [AMDGPU] gfx1010 SearchableTableEmitter patch for NSA.
Mon, May 13, 2:57 PM · Restricted Project
rampitec created D61875: [AMDGPU] gfx1010 clang target.
Mon, May 13, 2:56 PM · Restricted Project
rampitec added a reviewer for D61812: [AMDGPU] Fixed handling of imemdiate i1 literals: arsenm.
Mon, May 13, 2:02 PM · Restricted Project
rampitec created D61868: [AMDGPU] gfx1010 SearchableTableEmitter patch for NSA.
Mon, May 13, 12:39 PM · Restricted Project
rampitec committed rGd9930d499a55: [AMDGPU] gfx1010 tests. NFC. (authored by rampitec).
[AMDGPU] gfx1010 tests. NFC.
Mon, May 13, 12:29 PM
rampitec committed rL360615: [AMDGPU] gfx1010 tests. NFC..
[AMDGPU] gfx1010 tests. NFC.
Mon, May 13, 12:29 PM
rampitec committed rG79b2828b3f8f: [AMDGPU] Reorder includes per coding standard. NFC. (authored by rampitec).
[AMDGPU] Reorder includes per coding standard. NFC.
Mon, May 13, 11:04 AM
rampitec committed rL360609: [AMDGPU] Reorder includes per coding standard. NFC..
[AMDGPU] Reorder includes per coding standard. NFC.
Mon, May 13, 11:02 AM
rampitec committed rG21088639aed4: [AMDGPU] Remove now unused V2FP16_ONE constant def. NFC. (authored by rampitec).
[AMDGPU] Remove now unused V2FP16_ONE constant def. NFC.
Mon, May 13, 10:53 AM
rampitec committed rL360608: [AMDGPU] Remove now unused V2FP16_ONE constant def. NFC..
[AMDGPU] Remove now unused V2FP16_ONE constant def. NFC.
Mon, May 13, 10:53 AM
rampitec added a comment to D61812: [AMDGPU] Fixed handling of imemdiate i1 literals.

Why does this return false? A 1-bit immediate is either 0 or -1, both of which can be represented as inline constants everywhere.

Mon, May 13, 8:07 AM · Restricted Project

Sun, May 12

rampitec added inline comments to D61812: [AMDGPU] Fixed handling of imemdiate i1 literals.
Sun, May 12, 4:37 PM · Restricted Project

Fri, May 10

rampitec created D61812: [AMDGPU] Fixed handling of imemdiate i1 literals.
Fri, May 10, 4:43 PM · Restricted Project
rampitec accepted D61528: Revert "[AMDGPU] Enable code selection using `s_mul_hi_u32`/`s_mul_hi_i32`.".

LGTM

Fri, May 10, 3:01 PM

Thu, May 9

rampitec committed rG64196850f0e9: [AMDGPU] Pattern for v_xor3_b32 (authored by rampitec).
[AMDGPU] Pattern for v_xor3_b32
Thu, May 9, 5:07 PM
rampitec committed rL360395: [AMDGPU] Pattern for v_xor3_b32.
[AMDGPU] Pattern for v_xor3_b32
Thu, May 9, 5:06 PM
rampitec closed D61763: AMDGPU] Pattern for v_xor3_b32.
Thu, May 9, 5:06 PM · Restricted Project
rampitec created D61763: AMDGPU] Pattern for v_xor3_b32.
Thu, May 9, 3:22 PM · Restricted Project
rampitec committed rGa76da34b1d20: [AMDGPU] gfx1010 v_interp_* instructions (authored by rampitec).
[AMDGPU] gfx1010 v_interp_* instructions
Thu, May 9, 11:38 AM
rampitec committed rL360364: [AMDGPU] gfx1010 v_interp_* instructions.
[AMDGPU] gfx1010 v_interp_* instructions
Thu, May 9, 11:36 AM
rampitec closed D61703: [AMDGPU] gfx1010 v_interp_* instructions.
Thu, May 9, 11:36 AM · Restricted Project
rampitec committed rG4d4c9e0757c9: [AMDGPU] gfx1010 changes for PAL metadata (authored by rampitec).
[AMDGPU] gfx1010 changes for PAL metadata
Thu, May 9, 9:33 AM
rampitec committed rL360353: [AMDGPU] gfx1010 changes for PAL metadata.
[AMDGPU] gfx1010 changes for PAL metadata
Thu, May 9, 9:32 AM
rampitec closed D61704: [AMDGPU] gfx1010 changes for PAL metadata.
Thu, May 9, 9:32 AM · Restricted Project

Wed, May 8

rampitec committed rG327626368c5d: [AMDGPU] gfx1010 tests. NFC. (authored by rampitec).
[AMDGPU] gfx1010 tests. NFC.
Wed, May 8, 4:34 PM
rampitec committed rL360300: [AMDGPU] gfx1010 tests. NFC..
[AMDGPU] gfx1010 tests. NFC.
Wed, May 8, 4:34 PM
rampitec created D61704: [AMDGPU] gfx1010 changes for PAL metadata.
Wed, May 8, 3:54 PM · Restricted Project
rampitec added inline comments to D61528: Revert "[AMDGPU] Enable code selection using `s_mul_hi_u32`/`s_mul_hi_i32`.".
Wed, May 8, 2:59 PM
rampitec created D61703: [AMDGPU] gfx1010 v_interp_* instructions.
Wed, May 8, 2:48 PM · Restricted Project
rampitec accepted D61575: AMDGPU: Select VOP3 form of add.

LGTM

Wed, May 8, 2:27 PM
rampitec committed rG1dbf721315b8: [AMDGPU] gfx1010 exp modifications (authored by rampitec).
[AMDGPU] gfx1010 exp modifications
Wed, May 8, 2:22 PM
rampitec committed rL360287: [AMDGPU] gfx1010 exp modifications.
[AMDGPU] gfx1010 exp modifications
Wed, May 8, 2:21 PM
rampitec closed D61701: [AMDGPU] gfx1010 exp modifications.
Wed, May 8, 2:21 PM · Restricted Project
rampitec created D61701: [AMDGPU] gfx1010 exp modifications.
Wed, May 8, 2:17 PM · Restricted Project
rampitec added inline comments to D61575: AMDGPU: Select VOP3 form of add.
Wed, May 8, 11:57 AM

Tue, May 7

rampitec added inline comments to D61430: AMDGPU: Remove redundant patterns for shifts.
Tue, May 7, 3:42 PM
rampitec accepted D61359: [CodeGen] Rename DEBUG_TYPE for default hazard recognizer..

LGTM

Tue, May 7, 2:55 PM · Restricted Project
rampitec accepted D61564: [AMDGPU] Check MI bundles for hazards.

LGTM

Tue, May 7, 2:53 PM · Restricted Project
rampitec added a comment to D61564: [AMDGPU] Check MI bundles for hazards.

For the test itself, could you please add one more. You have test where hazard instruction follows bundle. Can you add the same, where it is in bundle, and another one precedes it? Thanks!

Sorry, I'm not sure I understand exactly what you mean. Do you want a hazard instruction in a bundle with another hazard instruction preceding it?

Tue, May 7, 2:07 PM · Restricted Project
rampitec added a comment to D61564: [AMDGPU] Check MI bundles for hazards.

The separate files helps me read which test is for which purpose

Tue, May 7, 8:40 AM · Restricted Project
rampitec added a comment to D61564: [AMDGPU] Check MI bundles for hazards.

There is no reason to split two test files.
For the test itself, could you please add one more. You have test where hazard instruction follows bundle. Can you add the same, where it is in bundle, and another one precedes it? Thanks!

Tue, May 7, 12:54 AM · Restricted Project

Mon, May 6

rampitec committed rG491746a58472: [AMDGPU] gfx1010 verifier changes (authored by rampitec).
[AMDGPU] gfx1010 verifier changes
Mon, May 6, 3:48 PM
rampitec committed rL360095: [AMDGPU] gfx1010 verifier changes.
[AMDGPU] gfx1010 verifier changes
Mon, May 6, 3:48 PM
rampitec closed D61521: [AMDGPU] gfx1010 verifier changes.
Mon, May 6, 3:48 PM · Restricted Project
rampitec committed rG971cb8b633b0: [AMDGPU] gfx1010: prefer V_MUL_LO_U32 over V_MUL_LO_I32 (authored by rampitec).
[AMDGPU] gfx1010: prefer V_MUL_LO_U32 over V_MUL_LO_I32
Mon, May 6, 3:25 PM
rampitec committed rL360094: [AMDGPU] gfx1010: prefer V_MUL_LO_U32 over V_MUL_LO_I32.
[AMDGPU] gfx1010: prefer V_MUL_LO_U32 over V_MUL_LO_I32
Mon, May 6, 3:25 PM
rampitec closed D61525: [AMDGPU] gfx1010: prefer V_MUL_LO_U32 over V_MUL_LO_I32.
Mon, May 6, 3:24 PM · Restricted Project
rampitec committed rG1bc001dec47e: [AMDGPU] gfx1010 memory legalizer (authored by rampitec).
[AMDGPU] gfx1010 memory legalizer
Mon, May 6, 2:56 PM
rampitec committed rL360087: [AMDGPU] gfx1010 memory legalizer.
[AMDGPU] gfx1010 memory legalizer
Mon, May 6, 2:54 PM
rampitec closed D61535: [AMDGPU] gfx1010 memory legalizer.
Mon, May 6, 2:54 PM · Restricted Project
rampitec added a comment to D61575: AMDGPU: Select VOP3 form of add.

I am in favor of this change in general, but can we fix folding issues before? We may have unwanted performance regressions otherwise.

I put a lot of time into trying, but fixing all of the issues will take time and this is an important workaround. The regression in the clmem lit test from increasing the folding threshold was worse. The folding pass needs more work to track VCC accurately, and the scheduler needs work to not regress it. An alternative might be to force shrinking of the addc

Mon, May 6, 1:18 PM
rampitec added a comment to D61575: AMDGPU: Select VOP3 form of add.

I am in favor of this change in general, but can we fix folding issues before? We may have unwanted performance regressions otherwise.

Mon, May 6, 12:34 PM